Our study followed a control–impact design, comparing rivers with natural and altered flow regimes in two natural (reference) flow regime types in northern Spain (Figs.  and ). classified hydrologically un-impacted Spanish rivers into 20 hydrological classes using data from 282 gauged naturally flowing rivers and relevant environmental and hydrological drivers such as climate, topography, land cover, and geology. Using these discriminators, they defined natural equivalent flow regimes (in the sense of estimated pre-impact reference flow regimes) for rivers that are nowadays impacted by various types of flow alteration . For our study, we selected 20 rivers within classes 13 and 10 of ; in each class, we sampled 4 naturally flowing rivers and 6 rivers with altered flow regimes (where one altered Mediterranean river had to be removed). Sampling sites were distributed within six basins, but we excluded locations on main stems and – except for one site – no site's catchment was nested within another one to maintain spatial independence (Fig. ). All sampling locations were second- to fourth-order rivers (Table S1 in the Supplement). For simplicity, the two (reference) flow regime types will be referred to as Atlantic and Mediterranean since class 13 is located in the Atlantic-type climatic region and class 10 in the Mediterranean-type climatic region of Spain, regardless of where these rivers discharge or which basin they belong to (Table , Fig. ). Overall, in our study, the sampled sites – referred to as rivers for simplicity – belong to one of four flow regimes: natural Atlantic (nA), altered Atlantic (aA), natural Mediterranean (nM), and altered Mediterranean (aM).

A natural Atlantic regime is characterized by the highest flows in early spring (February–April) due to snowmelt, followed by a gradual decrease in flow until summer low flows are reached and continue throughout July–September (Fig. a). Once minimum flows are reached in October, the flow rate picks up quite quickly, with a steep incline due to heavy rains in the fall. On the contrary, a natural Mediterranean regime (Fig. d) has its highest flows a bit later in the spring, in March/April, followed by a sharp decrease in flow in May consequently leading to a long, dry summer, reaching the minimum flows between August and September. As opposed to Atlantic rivers, the flow rate increases slowly leading back to the spring peaks, implying comparably less intense fall rains.

The altered rivers selected all have a dam with a reservoir (predominantly in the headwaters or upper reaches), although the exact alteration of the flow regime may vary depending on the size, purpose, and operation of the dam involved (Table ). The sampling sites in the altered rivers were located on average 4.1 km (range 1.74–33.6 km) downstream of a reservoir, and some sites had multiple upstream reservoirs. The catchment areas upstream of the sampling sites ranged from 89 to 1258 km2 (Table ), with an average of 429 km2. All the reservoirs in altered Mediterranean rivers were used for irrigation, while in altered Atlantic rivers, three reservoirs had irrigation and three reservoirs had hydropower production as their main purpose. Only one hydropower-producing reservoir, Narcea, has hydropeaking on a short timescale and has insufficient capacity for seasonal storage. On the other end of the spectrum, while Esla has the biggest reservoir capacity, Ebro has the highest storage index (i.e. highest capacity compared to its average annual runoff). In each river, we characterized DOM properties bi-monthly from October 2017 to August 2018, totalling six sampling occasions (October and December 2017 and February, April, June, and August 2018). Discharge measurements for each river and sampling day are given in Table S1.

We measured dissolved organic carbon (DOC) concentration, DOM absorbance and fluorescence, and DOM molecular size distribution. In the field, we filtered water samples through pre-combusted (450 °C, 4 h) glass fibre filters with a 0.3 μm nominal pore size (Advantec GF-75, Japan) into acid-washed, MilliQ-water-rinsed, pre-combusted (450 °C, 4 h) glass vials. We stored all the samples in the dark at 4 °C until analysis. We measured DOC concentration (mgL-1) by high-temperature catalytic combustion of an acidified and sparged water sample using a Shimadzu TOC-V Analyser with a typical limit of quantification (computed as 6 times the standard deviation of blanks) of <0.1±0.08 mgL-1 and a typical analytical precision of 1.5 %. To prepare the standards for the DOC measurements, we used potassium hydrogen phthalate (KHP). We took DOM absorbance and fluorescence measurements simultaneously using an Aqualog (Horiba Ltd, Kyoto, Japan). We used MilliQ water as a blank inserted every 10–15 measurements; each of the three field replicates was measured 3 times, and spectra were visually checked for consistency and outlier removal. We measured absorbance in 5 nm increments from 250 to 600 nm and fluorescence as excitation–emission matrices (EEMs); we excited the samples from 250 to 600 nm in 5 nm increments, with emission from 212 to 620 nm in 2 nm increments. We subtracted the MilliQ water EEMs from all sample EEMs to remove optical scatter phenomena but did not apply any inner filter correction since the decadal absorption coefficient at 254 nm (A254) was lower than the threshold of 0.3 for all samples .

We used the Aqualog measurements to calculate six optical indices: the fluorescence index (FI), the humification index (HI), the freshness index (β/α), the E2 : E3 ratio, the slope ratio (SR), and specific UV absorbance (SUVA₂₅₄), as summarized in Table . Further, we performed a parallel-factor analysis (PARAFAC) by setting the maximum emission to 1 after Raman normalizing, smoothing, and normalizing each EEM . We used the StaRdom package in R version 4.0.3 to partition the EEMs into individual components following the PARAFAC modelling strategy. We chose the eight-component PARAFAC model as the one achieving the best representation of the 699 EEM dataset. We verified the model using split-half analysis with random-sample selection following residual analysis. Sample-specific fluorescence of components was expressed in Raman units by reverting the normalization step through the multiplication of Fmax values with the original EEMs' maximum emission values. We scanned the literature via the OpenFluor database for similar components (Table ). The components C1, C2, C3, C4, and C5 were reported as old humic-like DOM with strong terrestrial signals composed of mostly high-molecular-weight (MW) substances. C6, C7, and C8 were all protein-like compounds of recently produced fresh material (Table ). We grouped the relative contribution of the first five components (C1–C5) into an indicator of humic and terrestrial signatures (Cterrestrial) by dividing the sum of the relative contributions by the sum of all components. In the same manner, we turned the last three components into an indicator of freshly produced microbial input (Cmicrobial). We used liquid size-exclusion chromatography (SEC) with organic carbon and nitrogen detection (LC-OCD-OND; ) for the molecular size distribution measurements. This allowed us to estimate the abundance of non-humic high-molecular-weight substances (CHMWS; like polysaccharides and proteins, mgCL-1); humic substances (CHS, mgCL-1); and neutral, hydrophilic-to-amphiphilic low-molecular-weight substances (CLMWS; e.g. aldehydes, sugars, and amino acids in mgCL-1). For samples below the detection limit, we used half the value of the detection limit in the final dataset to allow multivariate data analysis. To have an estimate of the relative contributions of each of the defined size fractions, we divided each by the total dissolved organic carbon concentration measured by SEC .

To assess differences among the flow regimes, we used principal component analysis (PCA) using the 81 flow indices reported in and grouped the rivers according to their flow regime to see if flow alterations homogenize or diversify natural flow characteristics (Fig. S1 in the Supplement). With this in mind, we also used the average distance to the centroid of a particular flow regime as a measure of dispersion in the same way as for DOM data (see below for the assessment of the among-river variation in annual mean DOM composition). The magnitudes of 1, 30, or 90 d high-flow events (1HF, 30HF, 90HF) and the magnitude of 1, 30, or 90 d low-flow events (1LF, 30LF, 90LF) were selected to compare the four flow regimes following the findings of natural and altered river flow differences from for the same rivers.

As measures of mean DOM concentration and mean composition for each river, we computed annual means of the DOC concentration, of optical indices, and of SEC fractions. To express the seasonal variability in these variables, we computed river-specific temporal coefficients of variation (CVs) and used them as indicators of annual shifts in DOM composition. Annual means and temporal CVs were then used as univariate responses describing DOM regimes to test for differences between the four flow regimes. For this, we followed a one-way approach in two steps for each response variable: first, we tested mean and CV differences in all four flow regimes with a one-way test , which – unlike classical ANOVA – does not need variance homogeneity as a prerequisite. In the case of significance, this was then followed by planned paired t tests between (i) natural Atlantic and natural Mediterranean (nA vs. nM), (ii) altered Mediterranean and natural Mediterranean (aM vs. nM), and (iii) altered Atlantic and natural Atlantic (aA vs. nA). We corrected the p values with the Bonferroni method, assuming three a priori planned tests. We also compared the spatial variation in annual means and temporal CVs across the rivers of the four flow regimes. For this, we followed a similar one-way approach in two steps, where we first tested for variance homogeneity of annual means and of temporal CVs across all four regimes with a Bartlett test, which – when significant – was followed by planned F tests between (i) nA vs. nM, (ii) aM vs. nM, and (iii) aA vs. nA. Taken together, our response variables included measures of mean location and temporal variation for each river's DOM indicators, which serve as DOM regime descriptors, and measures of (spatial) variation in these DOM regime descriptors across rivers belonging to the same flow regime. The latter was specifically computed to assess whether, in any given natural reference flow regime, dam-induced flow alteration diversifies DOM regimes across rivers compared to DOM regimes under natural flow.

Additionally, to assess the similarities and differences in annual average DOM composition and shifts in DOM composition between natural and altered rivers in a more integrative, multivariate way, we did a principal component analysis (PCA) using the relative fluorescence of PARAFAC components, optical indices, and SEC fractions. We looked for recognizable differences in DOM composition among flow regimes (i.e. between natural and altered regimes of both hydrological classes) with regard to location (mean DOM composition) and specifically with regard to temporal shifts in DOM composition in the various rivers. To statistically test for differences in DOM composition among flow regimes, we used a one-way permutational multivariate ANOVA on annual mean locations (river-specific centroids) in the PCA space with the same paired testing strategy as described above. Temporal shifts in DOM composition are graphically represented in 2D PCA space by polygons for each river (Fig. c–f), where each corner represents the DOM composition of a sampling date in the PCA space and the centroid represents the mean DOM composition of that river over the entire sampling period. The average distance of individual sampling dates to the centroid of a river serves as a measure of temporal shifts in DOM composition; it is computed as a dispersion (i.e. multivariate variation) on all dimensions of the PCA using the function betadisper() from the R package vegan . Here, a high average distance to the centroid (high dispersion) means strong temporal shifts in DOM composition. To statistically test the temporal shifts in DOM composition, we followed the same analysis approach as described above for a univariate response variable, i.e. a one-way ANOVA comparing dispersion as a response among all four flow regimes, followed by three planned t tests as post hoc tests.

Analogously to the analysis of univariate responses, we also compared the among-river variation in annual mean DOM composition and that of temporal shifts in DOM composition across the four flow regimes given by the multivariate dataset. Our motivation was again to assess whether, in any given natural reference flow regime, dam-induced flow alteration diversifies DOM regimes (in either mean location or temporal shifts in composition, i.e. dispersion) compared to DOM regimes under natural flow. Again, we followed a similar one-way approach in two steps. For the among-river variation in annual mean DOM composition, we tested for dispersion homogeneity of river centroids across all four regimes with PERMDISP, a permutational equivalent of the Bartlett test , which – when significant – was followed by planned permutational F tests between (i) nA vs. nM, (ii) aM vs. nM, and (iii) aA vs. nA. For the among-river variation in temporal shifts in DOM composition, we used the river-specific measure of temporal dispersion as a univariate response and followed the same approach as described above for variation in univariate responses (Bartlett test followed by planned t tests).

As the individual axes of the PCA can be understood as distinct meaningful components of a river's DOM regime, we also tested the differences in average location and variance along the PC1 and PC2 axes following the same univariate strategy of a one-way test followed by planned t tests, as described above. Similarly, we compared across-river variance of average location and variance along the PC1 and PC2 axes using a Bartlett test followed by planned F tests as post hoc analysis. A summary of the statistical workflow followed is included in Fig. .

To look for links between characteristics of the flow regime and temporal shifts in DOM composition, we performed three partial least-squares (PLSR) analyses (one for the overall shifts in DOM composition and one each for PC1 and PC2 variance). We chose PLSR analysis due to its reputed good performance in situations with high collinearity of predictors and small sample size . As predictors, we used the 81 hydrological indices computed by that were grouped into five flow regime components: magnitude, duration, frequency, timing, and rate of change (Table S3). As dependent variables, we used the multivariate dispersion (for the overall shifts in DOM composition) and the two variances on the first two PCA axes for each river, which can be understood as the two most important components of shifts in DOM composition, each having a specific qualitative character captured by the respective PCA axis. In the PLSR analysis, we selected the hydrological indices with VIP scores (variable importance on the predictor) higher than 1 as the most important explanatory variables and reported only the first component of the PLSR analysis for simplicity.

Both Atlantic and Mediterranean natural flow regimes are characterized by high flows in early to late spring and low flows throughout the summer, with a minimum in August–September before the fall rains begin (Fig. a and d). Additionally, natural Atlantic flow regimes have a second high-flow peak in early winter (November–December) compared to the natural Mediterranean regime.

On the contrary, altered Mediterranean rivers, with their irrigation reservoirs, do not experience the natural succession of high spring flows followed by summer droughts as observed in the natural Mediterranean rivers (Fig. d and e). From previous studies, we know that our altered Mediterranean rivers collect water during winter and spring and release it from May to September , which is the reason behind the high summer flows. We also know that this behaviour is limited by the storage capacity of these reservoirs, where a lower storage capacity requires an additional pre-release in late winter–spring due to the capacity being reached, resulting in a double-humped shape of the annual hydrograph (Fig. c and e).

When the mean of the altered Atlantic rivers is compared to its natural equivalent (Fig. a–c), we see that the abovementioned introduction of summer high flows is not as consistent; i.e. three rivers have distinct summer high flows, while three others have late-summer droughts. When grouped by reservoir purpose, rivers with hydropower reservoirs (Fig. b, Narcea, Nalon, Nansa) still show summer droughts even though they have altered flow regimes; i.e. they somewhat mimic the natural flow regime expected of that river. On the contrary, rivers with irrigation reservoirs (Fig. c, Porma, Esla, Arlanzon) have clear summer high flows and a secondary peak in spring due to limited storage capacity. Thus, in altered Atlantic rivers, flow regime heterogeneity among the rivers comes from the different purposes of the reservoirs (irrigation vs. hydropower), whereas in altered Mediterranean rivers, flow regime heterogeneity is derived mostly from the differences in storage capacity of reservoirs exclusively dedicated to irrigation.

From previous studies on our rivers, we know that altered rivers show a higher diversity of flow regimes compared to natural rivers . In a PCA analysis, we found that altered Mediterranean rivers have 13-fold and altered Atlantic rivers 44-fold higher dispersion in the PCA space compared to their natural equivalents, while both natural flow regimes showed similarly small dispersions (Fig. S1). This confirms that altered Atlantic regimes encompassing both irrigation and hydropower reservoirs are more diversified than altered Mediterranean regimes with irrigation reservoirs only.

As a final step, we looked at selected flow regime characteristics (Fig. ), specifically at the magnitudes of 1, 30, or 90 d high- and low-flow events (selected from ). These selected indicators are used to show the general trend of our flow regimes in magnitude and duration of annual minimum and maximum flows, covering short-duration flood pulses as well as seasonal droughts. Both natural flow regimes have higher 1 d high flows than their altered counterparts. Natural Mediterranean rivers have higher 30 d high flows than altered Mediterranean rivers, whereas this difference was not as clear between natural and altered Atlantic rivers. There was not much visible difference between regimes in 90 d high flows. Similarly, natural Atlantic rivers have lower 1 d low flows than altered Atlantic rivers with irrigation dams but do not have flows lower than on those with hydropower dams. Natural rivers have lower 30 d low flows and 90 d low flows than their altered counterparts for both Atlantic and Mediterranean rivers.

The DOC concentration varied 10-fold, from 0.59 to 5.3 mgL-1, across rivers throughout the year. Planned t tests did not show any significant difference in annual mean DOC concentration (Fig. a and Table ) between the two natural flow regimes (nA–nM) or between the altered and natural rivers (aA–nA and aM–nM, respectively). The consistently higher median in all altered Atlantic rivers compared to the natural Atlantic rivers and partly significant results prior to Bonferroni correction point to a lack of statistical power due to a low sample size, causing a lack of significance. The two highest average DOC concentrations belong to Duero and Ebro (Fig. a). We could not find any common hydrological characteristics separating these two from the remaining altered Mediterranean rivers, but note that Ebro has the highest storage index in the dataset. We also could not find any association of DOC concentration with discharge (Table S1). In fact, Ebro and Duero show completely opposite trends; Ebro has its highest DOC concentration in April at 1.05 m3s-1 and its lowest DOC value in August at 5.31 m3s-1, which coincides with the water release timings. On the contrary, Duero has its highest DOC concentration in June at 4.56 m3s-1 and its lowest DOC concentration in December at 0.88 m3s-1, when reservoirs are filled up.

The annual DOC variation in the rivers (CVDOC) ranged from 14 % to 53 %. The two natural regimes (nA–nM) showed no significant differences in CVDOC (Table ). Looking at the change in DOC through the sampled months in Atlantic rivers (Fig. b), we observe that both altered and natural rivers' average DOC time series follow similar paths throughout the year, with the lowest values in August followed by an increase until spring high flows. In Mediterranean rivers, however (Fig. c), altered rivers' DOC time series increase in April and June, while most natural rivers have low values in these months.

In both flow regimes, the among-river variation in DOC and CVDOC seems lower in natural rivers than in their altered equivalents, but we could not find any significant difference between any of the tested flow regime pairs (data not shown), likely again because of low statistical power.

In all regimes, DOM showed strong terrestrial characteristics throughout the year, with relatively low FI (<1.4) and β/α (<1) and high HI (0.86±0.04) and SUVA₂₅₄ (>2.4). Also, with values >60 %, the indicator Cterrestrial showed the dominance of soil-derived materials over microbially produced materials . The FI values ranged from 1.09 to 1.38 among the rivers throughout the year, typical of natural waters . The SR values were relatively low for all flow regimes (SR<1), indicating the dominance of higher-MW compounds and aromatic plant-derived materials . The relative contribution of SEC fractions also indicates that, in all of the regimes, humic substances were most abundant (CHS), followed by simple monomers like sugars and amino acids (CLMWS) and polysaccharides and proteins (CHMWS).

When we compared the two natural regimes of Atlantic and Mediterranean rivers, there was no significant difference in any annual average optical indicator or SEC fraction (Table ). However, CVSR and CVE2 : E3 were significantly higher in natural Atlantic rivers compared to natural Mediterranean rivers. The F-test results for all annual averages and compositional shift indicators show similar among-river variances between the two natural regimes (nA–nM; data not shown).

When we compared the natural regimes to their altered equivalents in terms of average DOM composition, we could not observe any significant difference in the annual average optical indices (Table ) in Atlantic or Mediterranean rivers. Among the CV values indicating shifts in DOM composition, we observed that in Atlantic rivers, CVE2 : E3, CVCprotein, and CVCterrestrial were significantly higher in natural rivers compared to their altered equivalents (Table ). In Mediterranean rivers, CVSR was significantly higher in altered rivers compared to the natural equivalents (could not be shown by the letters). From the F-test results, we observe that all of the average DOM composition and compositional shift indicators showed similar among-river variances between the natural and altered regimes (data not shown).

We further analysed the similarities and differences in the annual average and temporal shifts in DOM composition in a multivariate approach using a PCA. The first and second axes of the PCA (Fig. , PC1) explained 47 % and 16 % of the total variation in DOM composition. PC1 was negatively correlated with terrestrial and humic components (Cterrestrial), humic substances (CHS), and HI and was positively correlated with microbially sourced components (Cmicrobial), FI, β/α, and CHMWS, representing non-humic biopolymers (Huber et al., 2011). PC2 was negatively correlated with SR, E2 : E3 and SUVA₂₅₄ and positively correlated with CLMWS. Taken together, the PCA space from the bottom-left to the top-right corner shows a gradient of terrestrially sourced, aromatic, high-MW, and presumably diagenetically older material to microbially sourced, fresher, low-MW material. Notably, PC1 and PC2 capture indicators of DOM composition described by different methods: PC1 groups mostly fluorescence-based indices (Cmicrobial, Cterrestrial, FI, HI, β/α), and PC2 groups absorbance indices (E2 : E3, SUVA₂₅₄, SR) and SEC fractions. The patterns in PCA space were not related to DOC concentration or discharge (data not shown).

We used the centroid location and average distance to the centroid of each river (represented as polygons in 2D space in Fig. c–f) as proxies for the annual average and temporal dispersion of DOM composition, respectively. We could not find a significant difference in centroid location, i.e. annual average DOM composition, between any of the compared pairs of flow regimes (Table S2). In terms of temporal dispersions, we could not identify a significant difference between the natural rivers or between the natural and altered Mediterranean rivers, yet we observed that natural Atlantic rivers have a significantly higher average distance to the centroid compared to their altered Atlantic equivalents (Figs. c and d, a, and Table S2). Thus, in Atlantic rivers, DOM composition was more variable over time when the flow regime was natural. This effect was not as clear in Mediterranean rivers: although DOM regimes appeared to be less temporally variable in all altered Mediterranean rivers, they showed a higher diversity in natural rivers. This higher diversity of DOM regimes in natural Mediterranean rivers, with some being naturally quite dynamic while others show fairly stable DOM composition in time, precluded the identification of a significant flow alteration effect. Notably, however, Mediterranean rivers showed slightly more clear-cut flow alteration effects along the individual PCA axes (see below), with slightly more prominent excursions towards higher PC1 scores and lower PC2 scores in dam-impacted rivers towards late summer and fall.

The global tests comparing the four flow regimes did not show any significant differences in among-river variance regarding the mean centroid location and the mean dispersions from the river centroids (Table S2). Still, DOM regimes did not appear similarly variable among the various flow regimes; specifically, the seasonal shifts in DOM composition in natural Mediterranean rivers were quite variable. Here, by optical assessment, dams caused a homogenization across rivers with invariable DOM throughout the seasons (Fig. e and f). Also, seasonal shifts in DOM composition in natural Mediterranean regimes seemed to be more variable than in their Atlantic counterparts.

Since individual PCA axes represent qualitatively distinct shifts in DOM composition, we looked at the timelines, average locations, and variances in rivers along the PC1 and PC2 axes separately as well. Timelines of DOM composition throughout the year showed similar patterns on the PC1 axis for all flow regimes (Fig. S2a and b). On the contrary, on the PC2 axis, natural rivers differ from their altered equivalents in both August and October (Fig. S2c and d). Plotting the β/α index as a particular example, it turned out higher in October in altered Mediterranean rivers than in their natural equivalents; such a difference was not observed in other months or in Atlantic rivers (Fig. S2e and f). We could not find any significant difference in average location along the PC1 or PC2 axis between the two natural regimes or between the altered and natural Mediterranean rivers (Fig. b–c and Table S2). However, we can visually see that Atlantic rivers with irrigation dams have the three highest medians on the PC1 axis, while Atlantic rivers with hydropower dams have the three lowest medians.

We found that altered Atlantic rivers (Table S2) have lower average PC2 scores than natural Atlantic rivers. We could not find any significant difference in temporal variation along the PC1 or PC2 axis between the two natural rivers or between the altered and natural Mediterranean rivers (nM–aM; Fig. b–c and Table S2). However, we found that natural Atlantic flow regimes have significantly higher seasonal shifts in DOM composition than their altered equivalents along both the PC1 and PC2 axes (nA–aA; Table S2).

Moreover, along the PC1 axis, the among-river variation in mean DOM was similar among the rivers. Along the PC2 axis, we found a difference between the natural rivers (nA–nM; Table S2). Similarly, along PC1, among-river variation in seasonal shifts in DOM composition was similar for all the rivers. On the contrary, along the PC2 axis, it was different between both the natural rivers and the natural and altered Atlantic rivers (nA–nM, nA–aA; Table S2). These results confirm our visual assessments (Figs. c–f and ).

The first component of the partial least-squares regression (PLSR) models explained 30.9 % of the variance in the dispersion in the entire multidimensional PCA space, 18.7 % of the variance on the PC1 axis, and 24.0 % of the variance on the PC2 axis. For ease of interpretation, only the first component was selected from each of the three PLSR models. Based on VIP >1 as a criterion, 34, 34, and 29 indices were selected to explain the multidimensional dispersion and the variance in the PC1 and PC2 axes, respectively. Mostly, indices identified as important for “multidimensional dispersion” – as the global model of shifts in DOM composition – were then also found to be relevant for explaining variances in PC1, PC2, or both axes. A few indices, however, were identified as relevant in only one of the three models, suggesting that multivariate dispersion may be a noisy but more comprehensive measure of shifts in DOM composition, i.e. capturing shifts in DOM composition that cannot be attributed solely to either PC1 or PC2.

The model for dispersion was driven by high flow in the wet winter and spring months (November, December, and January flow and March, April, and May flow; Fig. a) and low flow in the dry summer months (June, July, August, and September flows; Fig. a). Standard deviations of these flow magnitude indices showed a similar trend of positive correlation in wet months and negative correlation in dry months. Moreover, the dispersion increased with increasing magnitude and frequency of short-duration high-flow events (1, 3, or 7 d maximum and 1, 3, or 7 d frequency; Fig. a). Regarding the rate and frequency of flow change, the count of increasing flow days decreased and the count of decreasing flow days increased this dispersion (no increasing flow days, no decreasing flow days; Fig. a). Regarding the timing indices, the later date of the annual minimum flow was related to higher dispersion, whereas variation in this date (i.e. its standard deviation) was related to lower dispersion.

In addition to identifying the drivers of dispersion, we observed that PC2 variance decreased with increasing long-duration minimum flows, specifically while PC1 variance increased with increasing high flood frequency and counts. Similarly, PC1 variance increased with increasing rise rate and decreased with increasing fall rate. Additionally, we found that PC1 variance decreased with increasing SD of minimum flood-related indices, especially when most SD indices were negatively correlated.

This study benefits from the work of , who classified natural flow regimes on the Iberian Peninsula into 20 distinct regime types (or hydrological classes), of which we selected 2. Indeed, natural rivers in both classes have pretty distinct flow regimes, each with common features. Hence, both natural flow regimes have a low variance of flow descriptors among their rivers (Fig. 1c). On the contrary, the high among-river variations in flow variability boxplots in most altered regimes (Fig. ) combined with the hydrographs (Fig. ) indicate little commonality in flow regimes across the altered rivers: while some rivers have smoother hydrographs, others show a tendency towards increased temporal variation in their flow regimes (e.g. introducing summer high flows and/or winter low flows) or show little influence of the dam on the seasonal timescale studied. This is in accordance with the literature suggestions of diverse effects upon damming, and while it may push some rivers outside of their normal hydrological functioning , others may still behave similarly to their expected natural flow regimes.

As visible by their higher among-river variation, flow alterations diversified Atlantic rivers more than Mediterranean rivers (Fig. S3). Indeed, both reference flow regimes diversified in different directions, mostly due to various reservoir designs, purposes, and sizes. Altered flow regimes are thus not as easily classified and not as distinct as their natural equivalents. However, we also observed two common features of altered flow regimes: (1) the general loss of dynamics throughout the year through less intense 1 d high flows (Fig. a) and more intense seasonal low flows (Fig. e and f) and (2) the newly introduced summer high flows resulting in a shift in flow seasonality (Fig. b and e). The latter was especially pronounced in Mediterranean rivers, where reservoirs exclusively serve the purpose of irrigation and thus release water to arid downstream regions during summer. Additionally, the reservoir capacity is surpassed in some but not all smaller reservoirs in late winter to early spring, resulting in an unreliably appearing secondary earlier “release”, hence the double-humped shape of some altered flow regimes. Reservoirs in our study ranged in size between 2 and 664 hm3; thus some but not all were big enough to hold high spring flows for storage until release in a single summer peak. A similar seasonal water release does not exist in the altered Atlantic rivers, especially those flowing to the north (Nalon, Nanca, Narcea). These hydropower reservoirs have low storage capacity; hence – at the seasonal timescale studied – they more closely mimic the intense summer low flows of their natural equivalents. We note that hydropower as a purpose does not in principle prevent seasonal water storage for later hydropeaking , yet hydropeaking was not observed in the studied rivers.

Thus, in our study, flow regimes diversify through common yet varied types of alteration. Our study design implied the creation of inherently heterogeneous altered flow regime groups, with reservoirs differing in size and residence time, but also facilitated various stratification patterns and showed various kinds of warming potential. One thing our study cannot achieve by design is to look for commonalities in DOM regimes for reasonably categorized, distinct types of flow alteration. In fact, grouping reservoirs by purpose (irrigation vs. hydropower) resulted in little difference between the studied DOM regimes despite their distinct flow features. Thus, we believe that any commonality we find based on a coarser definition of mere “dam-induced flow alteration” may be of prominent importance and strong applicability.

In our comparison of annual averages of DOM quantity (DOC) in natural and altered rivers, tests missed significance likely due to low statistical power despite median values of altered rivers appearing higher, specifically in the Atlantic rivers. Yet, a clear-cut response of DOM quantity to flow regimes or their alteration is maybe naive to expect. The DOM quantity of natural rivers is affected by various environmental factors such as topography, climate, and land cover. In contrast, in altered rivers, it is additionally influenced by factors such as dam operation, reservoir stratification, and reservoir water temperature . Various studies have found conflicting results for reservoirs: increasing, decreasing, or not altering DOC concentrations of the downstream systems . In rivers from arid regions with high flow variation, for example, upstream reservoirs may act as a carbon source or a carbon sink . In temperate regions, upstream reservoirs may act as a carbon source , while no change was observed in the reservoirs of boreal rivers .

The existing diversity of influences makes a common response of DOM quantity to flow alteration across Atlantic and Mediterranean rivers unlikely, especially when considering the highly integrative nature of a variable like annually averaged DOC that integrates DOM across various sources and chemical compositions, as well as over time. Yet, in our study, we observed an increasing trend in DOC from natural to altered rivers when Atlantic and Mediterranean rivers were separated. One explanation for this apparent effect of dams on DOC concentration could reflect a sampling problem associated with the efficient capture, transient storage, and averaging of extreme DOM amounts delivered to the river network from the catchment during short-term flood events . In this explanation, dams with large enough reservoirs interfere with the downstream transmission of a DOM pulse during a flood, but more importantly, this DOM pulse is transiently stored in the reservoir and – if not subjected too strongly to bio- and photodegradation – passed on downstream later over a longer time period. Thanks to dams interfering with naturally occurring flow and DOM pulses, downstream DOM composition is “hydrologically averaged” and invariant in time. Thus, sampling at discrete points in time in altered, downstream rivers results in relatively high and stable values, while it is highly unlikely that with just six sampling dates we were able to capture a pulse on any of our natural rivers, whose annual mean DOC then ends up being underestimated. In addition or as an alternative to this effect of temporal smoothing of terrigenous DOM pulses, productivity in reservoirs may contribute additional DOM of autochthonous origin, which could potentially also counteract losses by bio- and photodegradation. All these mechanisms are associated with the residence time of water in the reservoirs, yet they expectedly result in different DOM compositions passed on to downstream river reaches.

Strictly looking at annual average measures for DOM composition between natural and altered flow regimes, we found only limited evidence for differences; we observed a significant flow alteration effect on mean DOM composition only along PC2 for Atlantic rivers. Here, altered rivers appeared on average more humic, with signs of photodegradation (SR). The (further) lack of significant effects between naturally flowing and flow-altered rivers with regard to (average) DOM composition to some extent supports the idea of reservoirs as smoothers of a variable, natural DOM regime, or it may imply that other factors such as hydrology or catchment characteristics have an overriding influence on DOM composition, thereby masking any potential reservoir effect by a lot of noise.

In fact, the literature on linking DOM composition to dams has mixed results. Some studies show that a high export of autochthonous and biolabile DOM caused by increased primary production in reservoirs and tailwaters may result in a shift from terrestrially sourced DOM to protein-like DOM downstream of dams . On the contrary, other studies claim that dams of certain sizes and functions may have no significant effect on downstream DOM composition. In boreal rivers, for example, no shift in DOM composition from terrestrial to protein-like was observed downstream of hydropower and storage dams . However, another study from the same climatic region observed that increased residence time in a hydropower-producing mesotrophic reservoir shifted the DOM composition from allochthonous to autochthonous . This diversity of responses again points to a particular source of noise in our study that may have caused difficulties in finding clear-cut effects of flow alteration on DOM composition: the fact that we grouped various dams of different purposes and sizes under the term “flow alteration”. In this respect, reservoir trophic state and dam operations, such as the timing of the reservoir filling or water release, are also of concern, specifically regarding their influence on the composition of DOM transported to downstream reaches by interfering with residence times and the transmission of upstream source signals . Indeed, dams can be a carbon source or a carbon sink depending on many such factors related to the dam's nature and operation; potentially, they cause DOM composition to respond to flow alteration in opposite ways.

A DOM regime may be better described than just by annually averaged DOC and DOM composition, especially when these variables are subject to dynamic processes and show pronounced temporal variability. Indeed, when looking at the temporal change in DOM composition indicators (Fig. S2), we observed specific differences in summer and fall months between the natural and altered regimes in both Atlantic and Mediterranean rivers. The divergence between natural and altered rivers is even more obvious in the multivariate analysis, specifically along PC2 as well as in β/α, where in particular during summer, DOM in altered rivers behaved differently than in natural rivers (Fig. S2c–f).

To make use of the temporally spaced information without having to resort to time series analyses, we computed measures of variation in DOC and of shifts in DOM composition on the annual timescale as descriptors of the DOM regime. We reasoned that such integrative measures based on variance can also be quite flexibly computed in cases where the temporal spacing of measurements is uneven, where it is not aligned among rivers, or where sampling is sparse or spread randomly over multiple years.

Hydrological seasonality and event-induced flow variation are the main drivers of annual variation in carbon concentration in natural rivers . However, we could not observe an effect of flow regime on the annual DOC variation, nor could we observe any clear differences among annual timelines of DOC (Fig. b and c). On the contrary, we found several DOM indices showing differences in their annual variance (CV values) and in multivariate dispersion among natural and altered flow regimes. Specifically, dam-impacted Atlantic rivers showed consistently lower temporal variation in various indices, multivariate dispersion, and PCA axes. Thus, dams in these rivers led to a significant temporal homogenization of DOM in downstream sections. These patterns appeared in Mediterranean rivers as well, most notably when assessing multivariate dispersion (Fig. 6). However, in Mediterranean rivers, the differences between natural and altered flow regimes were insignificant for most indicators, largely because of remarkably higher spatial variation in temporal CVs across the various natural Mediterranean rivers. DOM regimes in naturally flowing Mediterranean rivers were indeed either quite variable over time or fairly stable – a finding that strongly contrasts with the consistently dynamic DOM regimes of natural Atlantic rivers (Figs. 6 and 7). Taken together, our results suggest that dams may lead to fairly consistent and temporally stable DOM regimes in downstream river sections even in climatically different reference flow regimes and contrast the high diversity of DOM regimes potentially appearing in natural flow, especially in Mediterranean rivers.

Our PLSR results linking flow regime characteristics to multidimensional dispersion and variance along the PC1 and PC2 axes yield more insights into the mechanisms behind seasonal shifts in DOM composition. Short-duration high-flow events are the key to moving terrestrially stored carbon into the riverine system via increased riparian connectivity. The PLSR results show wider annual shifts in DOM composition, with an increase in indices describing magnitude (1, 3, and 7 d maximum) and frequency (1, 3, and 7 d flood frequency) of such short-duration high-flow events (Fig. a). These relationships may be partly driven by natural differences in terrestrial DOM mobilization, but it is likely that they are also associated with the ability of reservoirs to interfere with the downstream transmission of a terrigenous DOM pulse.

Indeed, many reservoirs in northern Spain are associated with a decreased frequency of high-flow events due to the storage of water in the reservoir and downstream flow smoothing . In our study, in both natural flow regimes, summer low flows follow high winter and spring flows (with some climate-specific variation, for instance in the exact timing and intensity of drought). Thus, flow alteration largely leads to a more-or-less strong reversal of seasonal dynamics by retaining water in reservoirs in winter and spring for later release during summer. This effect is greater in Mediterranean rivers. This seasonality of the flow regimes in our study is reflected in patterns of shifting DOM composition: PLSR modelling identified a positive effect of mean monthly flows in winter and spring (floods) but a negative effect of mean monthly flows in summer (droughts) on the magnitude of the shifts in DOM composition. This is corroborated by relatively higher terrestrial signals in the rainy spring months in natural rivers (Fig. S2a and b) and also shown by the positive effect of seasonal high flows (March, April, May) on PC1, indicating aromaticity and molecular weight variance. This is in accordance with the current scientific consensus that high flows are generally linked to allochthonous DOM inputs, while low flows promote autochthonous production . Natural flow regimes allow the transmission of flood-induced terrigenous DOM pulses in spring while such DOM is rare in these rivers during summer low flows, when there is increased opportunity for production of autochthonous DOM. On the other hand, in altered flow regimes, reservoirs pass on seasonally smoothed and thus rather invariable terrigenous DOM to downstream systems long into summer (Fig. S2e and f). The negative effect of summer flows on the magnitude of shifts in DOM composition is likely as much driven by low summer flows in natural flow regimes as by high summer flows in altered flow regimes. In altered flow below dams, the lack of summer drought lowers the natural variation in the DOM composition of a river.

On the other hand, low flow during naturally dry summer months may directly increase the variation in DOM composition through higher microbial signals from autochthonous production of DOM in the barely flowing rivers. This is shown by higher PC1 and PC2 values during summer low flow in the natural Atlantic rivers, while DOM remained comparably invariable in the altered Atlantic rivers (Fig. S2). A similar microbially sourced carbon pulse was observable in altered rivers in October, coinciding with the lowest flow rate for these rivers. This finding still agrees with the earlier interpretation of reservoirs as hydrological smoothers of upstream DOM signals, which, however, may include not only terrigenous DOM but also material produced in stream. Within reservoirs, these two types of material may then be subjected to photo- and bio-degradation, ultimately leaving less labile humic material for continuous downstream export with little compositional variation.

A counter-effect of the lowered autochthonous contribution from upstream rivers to DOM variance in the altered rivers may be the potential production of DOM in the reservoirs, especially when reservoirs have longer residence times , have a reasonable nutrient supply , and support warming. For Atlantic rivers, our data do not suggest reservoirs as sources of new autochthonous material. For Mediterranean rivers, summertime excursions of DOM towards higher PC1 scores (i.e. towards more microbial DOM and higher FI and β/α ratios; Fig. S2b, d, and f) suggest some influence of DOM production within reservoirs, even though this effect was not statistically tested. These results agree with the generally good ecological status and thus low trophic state of the reservoirs (Table 1).

Notably, the buffering effect of reservoirs may also imply a change in DOM composition due to higher degradation as a result of longer residence time. We observed that in the August–October measurements, PC2, which is strongly correlated with SR and E2 : E3, diverges between natural and altered rivers in both the Atlantic and Mediterranean regimes. Here, natural rivers have positive values, and altered rivers have negative values. We think that this may be related to the higher irradiation, as SR indicates photodegradation. This might be due to two factors: longer residence times in the reservoir leading up to these moments of sampling and water release from the reservoirs over the summer that decreases water depth and thus increasing light penetration.

The pervasive effects of dams on DOM regimes also had an interesting additional effect, when looking at the spatial variation in DOM regimes across rivers. When assessing DOM composition, we observed that flow alterations homogenized the DOM regime across rivers despite our dams being of differing sizes and purposes, altering flow regimes in various directions (e.g. Fig. ). This loss of the natural spatial diversity of DOM regimes across rivers indicates that dams have a powerful homogenizing effect on DOM regimes at the landscape scale. Interestingly, we could not find such an effect for the variation in DOC and CVDOC. Here, data inspection (Fig.  and Table ) shows a non-significant trend towards increasing spatial variation across rivers, from natural to altered rivers. This would indicate a spatial diversification of DOM regimes in terms of quantitative behaviour, which parallels the flow regime diversification due to dams.